What happened. A Pactum requisition deployment remained in shadow mode until the agent exceeded 95% accuracy against the customer’s existing human-review process, before moving into live adviser mode.
What it actually means. Autonomy should not begin with permission. It should begin with evidence.
Shadow mode lets an agent work on live inputs without affecting the live process. Its output can be compared against human reviewers, edge cases can be corrected and the organisation can define the quality threshold required before the agent influences real work.
This is the gap most demos skip. A convincing output is not the same as repeatable production performance.
Trust is earned through deployment data.
What to watch. Whether shadow, adviser and autonomous modes become a standard maturity path for procurement-agent implementations.
Disclosure. Daniel works at Pactum. The editorial view is his own.